promptlyzer 1.5.1

Official Python SDK for Promptlyzer - Multi-model routing layer for LLMs

Promptlyzer Python SDK

Official Python client for Promptlyzer - Multi-model routing layer for LLMs.

---

What is Promptlyzer?

Promptlyzer is a production-ready LLM infrastructure layer that provides:

• Smart model routing with complexity-based deployment strategies
• Multi-provider support (OpenAI, Anthropic, Together AI, SiliconFlow)
• Automatic cost optimization and credit management
• Prompt management with versioning and caching
• Real-time analytics and monitoring

---

Installation

pip install promptlyzer

# Upgrade to latest
pip install --upgrade promptlyzer

Requirements:

• Python 3.7+
• Promptlyzer API key

---

Quick Start

Step 1: Setup Your Account

1. Sign up at promptlyzer.com
2. Create a team (or use the default team created for you)
3. Create a project to organize your deployments
4. Get your API key from Settings → API Keys
5. (Optional) Add your provider API keys at Providers page to use your own quota instead of credits

Step 2: Install and Initialize

pip install promptlyzer

# Set environment variable
export PROMPTLYZER_API_KEY="pk_live_YOUR_API_KEY"

Step 3: Create a Deployment

Why use deployments?

• AI-powered smart routing across multiple model tiers
• Automatic cost optimization based on request complexity
• Real-time monitoring and analytics
• No code changes needed when switching models

Create a deployment from your dashboard:

1. Go to Deployments page in your project
2. Click "Create Deployment"
3. Configure your 3-tier routing:
   • Tier 1 (Fast & Cheap): e.g., gpt-4o-mini - Simple queries
   • Tier 2 (Balanced): e.g., gpt-4o - Standard requests
   • Tier 3 (Premium): e.g., claude-sonnet-4-5 - Complex tasks
4. Copy your deployment ID

Smart Routing Engine:

• Uses AI to analyze prompt complexity (0-100 score)
• Considers task type and context length
• Routes to optimal tier automatically

Step 4: Use Deployment for Inference

from promptlyzer import PromptlyzerClient

# Initialize client
client = PromptlyzerClient(api_key="pk_live_YOUR_API_KEY")

# Make inference request through deployment
response = client.deployments.infer(
    deployment_id="dep_abc123",
    prompt="Explain quantum computing in simple terms",
    task_type="customer_agent",  # Optional: helps routing decision
    context="Previous conversation..."  # Optional: chat history or documents
)

# Response contains
print(f"Response: {response['response']}")
print(f"Model: {response['model_used']}")
print(f"Tier: {response['tier']}")
print(f"Cost: ${response['cost_usd']:.6f}")
print(f"Latency: {response['latency_ms']}ms")

Smart Routing Logic (AI-Powered):

• Analyzes prompt + context + task type
• Calculates complexity score (0-100)
• Routes to appropriate tier:
  • Tier 1 (0-40): Simple FAQ, quick responses
  • Tier 2 (40-70): Standard queries, moderate context
  • Tier 3 (70-100): Complex reasoning, large context

Using Your Own API Keys (No Credits):

response = client.deployments.infer(
    deployment_id="dep_abc123",
    prompt="Your prompt here",
    provider_api_key="sk-..."  # Your OpenAI/Anthropic key
)

View Deployment Logs

# Get all logs for a deployment
logs = client.deployments.get_logs(
    deployment_id="dep_abc123",
    status="success",  # or "error"
    limit=50
)

for log in logs:
    print(f"{log['timestamp']}: {log['model_id']} - ${log['cost_usd']:.6f}")

Alternative: Direct Inference (Without Deployment)

Use the same client for direct inference:

# Simple inference using credits (no deployment)
response = client.inference.infer(
    prompt="What is machine learning?",
    model="gpt-4o-mini"
)

print(response.content)
print(f"Cost: ${response.usage.cost_usd:.6f}")

One client, multiple features:

• client.deployments.infer() → Smart routing with deployments
• client.inference.infer() → Direct model inference
• client.get_prompt() → Prompt management
• client.billing.get_balance() → Credit management

---

Supported Models

OpenAI

• gpt-5
• gpt-4o (vision supported)
• gpt-4o-mini
• gpt-3.5-turbo

Anthropic

• claude-3-5-sonnet-20241022 (vision supported)
• claude-3-haiku-20240307
• claude-3-opus-20240229

Together AI

Llama Models:

• llama-3.3-70b-turbo
• llama-3.3-70b-instruct-turbo
• llama-3.1-8b-turbo
• llama-3.2-3b
• llama-4-scout (vision supported)
• llama-4-maverick (vision supported)

Qwen Models:

• qwen-2.5-72b
• qwen-2.5-7b
• qwen-qwq-32b
• qwen3-235b

DeepSeek Models:

• deepseek-v3
• deepseek-v3.1
• deepseek-r1-0528 (reasoning model)
• deepseek-r1-distill-llama-70b
• deepseek-r1-distill-qwen-14b
• deepseek-r1-distill-qwen-1.5b

Other Models:

• mixtral-8x7b
• kimi-k2-instruct
• openai-gpt-oss-120b
• openai-gpt-oss-20b

SiliconFlow

Vision Models:

• zai-org/GLM-4.5V
• stepfun-ai/step3
• THUDM/GLM-4.1V-9B-Thinking (thinking model)
• Qwen/Qwen2.5-VL-72B-Instruct
• Qwen/Qwen2.5-VL-32B-Instruct
• Qwen/Qwen2.5-VL-7B-Instruct
• deepseek-ai/deepseek-vl2

Text Models:

• deepseek-ai/DeepSeek-V3.1
• deepseek-ai/DeepSeek-R1 (reasoning model)
• deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
• zai-org/GLM-4.5
• zai-org/GLM-4.5-Air
• Qwen/Qwen3-235B-A22B-Instruct-2507
• Qwen/Qwen3-235B-A22B-Thinking-2507 (thinking model)
• moonshotai/Kimi-K2-Instruct
• tencent/Hunyuan-A13B-Instruct
• Qwen/QwQ-32B
• Qwen/Qwen2.5-72B-Instruct
• Qwen/Qwen2.5-32B-Instruct
• Qwen/Qwen2.5-7B-Instruct
• meta-llama/Meta-Llama-3.1-8B-Instruct

View pricing and details: Model Explorer

---

Authentication

Method 1: Environment Variable (Recommended)

export PROMPTLYZER_API_KEY="pk_live_your_api_key"

from promptlyzer import PromptlyzerClient
client = PromptlyzerClient()  # Auto-loads from environment

Method 2: Direct Initialization

client = PromptlyzerClient(api_key="pk_live_your_api_key")

---

Billing & Credits

Check Balance

balance = client.billing.get_balance()
print(f"Available: ${balance['available']:.2f}")
print(f"Used: ${balance['used']:.2f}")

---

Prompt Management

Get Prompt

prompt = client.get_prompt(
    project_id="proj_123",
    prompt_name="customer_support",
    environment="dev"  # or "staging", "prod"
)

print(prompt['content'])

Use Prompt in Inference

response = client.inference.infer(
    prompt=prompt['content'],
    model="claude-3-5-sonnet"
)

Prompts are cached for 5 minutes to reduce API calls.

---

Advanced Features

Streaming Responses

for chunk in client.inference.infer(
    prompt="Write a story",
    model="gpt-4o",
    stream=True
):
    print(chunk.content, end="", flush=True)

Error Handling

from promptlyzer.exceptions import (
    AuthenticationError,
    InsufficientCreditsError,
    RateLimitError,
    InferenceError
)

try:
    response = client.inference.infer(prompt, model)
except AuthenticationError:
    print("Invalid API key")
except InsufficientCreditsError as e:
    print(f"Out of credits: {e.available_credits}")
except RateLimitError as e:
    print(f"Rate limited. Retry after: {e.retry_after}s")
except InferenceError as e:
    print(f"Inference failed: {e.message}")

---

Configuration

Unified Client

from promptlyzer import PromptlyzerClient

client = PromptlyzerClient(
    api_key="pk_live_YOUR_API_KEY",
    environment="dev"  # Options: "dev", "staging", "prod" (default: "dev")
)

One client for everything:

• client.deployments.* → Deployment inference with smart routing
• client.inference.* → Direct model inference
• client.get_prompt() → Prompt management across environments
• client.billing.* → Credit and usage management

Internal Settings (auto-configured):

• Prompt cache TTL: 5 minutes
• Connection pool size: 10
• Request timeout: 10s (quick), 60s (inference)
• Retry strategy: 3 attempts with exponential backoff

---

Documentation

Full documentation: DOCUMENTATION.md

Topics covered:

• Deployment configuration and usage
• Advanced inference options
• Prompt management and versioning
• Cost analytics and monitoring
• Production best practices

---

Support

• Documentation: docs.promptlyzer.com
• Dashboard: promptlyzer.com
• Email: [e-mail]

---

License

MIT License - See LICENSE for details.

---